Updates to prepare for Apriel 1.6 release

app.py

@@ -30,7 +30,12 @@ BUTTON_WIDTH = 160
 DEFAULT_OPT_OUT_VALUE = DEBUG_MODE
 
 # If DEBUG_MODEL is True, use an alternative model (without reasoning) for testing
-DEFAULT_MODEL_NAME = "Apriel-1.5-15B-thinker" if not DEBUG_MODEL else "Apriel-1.5-15B-thinker"  # "Apriel-5b"
+DEFAULT_MODEL_NAME = "Apriel-1.6-15B-Thinker" if not DEBUG_MODEL else "Apriel-1.6-15B-Thinker"
+
+SHOW_BANNER = False
+BANNER_MARKDOWN = """
+                    <span class="banner-message-text">ℹ️ This app has been updated to use the recommended temperature of 0.6. We had set it to 0.8 earlier and expect 0.6 to be better. Please provide feedback using the model link.</span>
+                    """
 
 BUTTON_ENABLED = gr.update(interactive=True)
 BUTTON_DISABLED = gr.update(interactive=False)
@@ -126,6 +131,9 @@ def run_chat_inference(history, message, state):
     error = None
     model_name = model_config.get('MODEL_NAME')
     temperature = model_config.get('TEMPERATURE', DEFAULT_MODEL_TEMPERATURE)
+    output_tag_start = model_config.get('OUTPUT_TAG_START', "[BEGIN FINAL RESPONSE]")
+    output_tag_end = model_config.get('OUTPUT_TAG_END', "[END FINAL RESPONSE]")
+    output_stop_token = model_config.get('OUTPUT_STOP_TOKEN', "<|end|>")
 
     # Reinitialize the OpenAI client with a random endpoint from the list
     setup_model(model_config.get('MODEL_KEY'))
@@ -435,19 +443,19 @@ def run_chat_inference(history, message, state):
             output += reasoning_content + content
 
             if is_reasoning:
-                parts = output.split("[BEGIN FINAL RESPONSE]")
+                parts = output.split(output_tag_start)
 
                 if len(parts) > 1:
-                    if parts[1].endswith("[END FINAL RESPONSE]"):
-                        parts[1] = parts[1].replace("[END FINAL RESPONSE]", "")
-                    if parts[1].endswith("[END FINAL RESPONSE]\n<|end|>"):
-                        parts[1] = parts[1].replace("[END FINAL RESPONSE]\n<|end|>", "")
-                    if parts[1].endswith("[END FINAL RESPONSE]\n<|end|>\n"):
-                        parts[1] = parts[1].replace("[END FINAL RESPONSE]\n<|end|>\n", "")
-                    if parts[1].endswith("<|end|>"):
-                        parts[1] = parts[1].replace("<|end|>", "")
-                    if parts[1].endswith("<|end|>\n"):
-                        parts[1] = parts[1].replace("<|end|>\n", "")
+                    if parts[1].endswith(output_tag_end):
+                        parts[1] = parts[1].replace(output_tag_end, "")
+                    if parts[1].endswith(f"{output_tag_end}\n{output_stop_token}"):
+                        parts[1] = parts[1].replace(f"{output_tag_end}\n{output_stop_token}", "")
+                    if parts[1].endswith(f"{output_tag_end}\n{output_stop_token}\n"):
+                        parts[1] = parts[1].replace(f"{output_tag_end}\n{output_stop_token}\n", "")
+                    if parts[1].endswith(f"{output_stop_token}"):
+                        parts[1] = parts[1].replace(f"{output_stop_token}", "")
+                    if parts[1].endswith(f"{output_stop_token}\n"):
+                        parts[1] = parts[1].replace(f"{output_stop_token}\n", "")
 
                 history[-1 if not completion_started else -2] = gr.ChatMessage(
                     role="assistant",
@@ -528,14 +536,11 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
             }}
         </style>
         """, elem_classes="css-styles")
-    with gr.Row(variant="compact", elem_classes=["responsive-row", "no-padding"], ):
-        with gr.Column():
-            gr.Markdown(
-                """
-                <span class="banner-message-text">ℹ️ This app has been updated to use the recommended temperature of 0.6. We had set it to 0.8 earlier and expect 0.6 to be better. Please provide feedback using the model link.</span>
-                """
-                , elem_classes="banner-message"
-            )
+    if SHOW_BANNER:
+        with gr.Row(variant="compact", elem_classes=["responsive-row", "no-padding"], ):
+            with gr.Column():
+                gr.Markdown(BANNER_MARKDOWN, elem_classes="banner-message")
+
     with gr.Row(variant="panel", elem_classes="responsive-row"):
         with gr.Column(scale=1, min_width=400, elem_classes="model-dropdown-container"):
             model_dropdown = gr.Dropdown(

utils.py

@@ -12,6 +12,20 @@ DEBUG_MODE = False or os.environ.get("DEBUG_MODE") == "True"
 DEBUG_MODEL = False or os.environ.get("DEBUG_MODEL") == "True"
 
 models_config = {
+    # "Apriel-1.6-15B-Thinker": {
+    #     "MODEL_DISPLAY_NAME": "Apriel-1.6-15B-Thinker",
+    #     "MODEL_HF_URL": "https://huggingface.co/ServiceNow-AI/Apriel-1.6-15b-Thinker",
+    #     "MODEL_NAME": os.environ.get("MODEL_NAME_APRIEL_1_6_15B"),
+    #     "VLLM_API_URL": os.environ.get("VLLM_API_URL_APRIEL_1_6_15B"),
+    #     "VLLM_API_URL_LIST": os.environ.get("VLLM_API_URL_LIST_APRIEL_1_6_15B"),
+    #     "AUTH_TOKEN": os.environ.get("AUTH_TOKEN"),
+    #     "REASONING": True,
+    #     "MULTIMODAL": True,
+    #     "TEMPERATURE": 0.6,
+    #     "OUTPUT_TAG_START": "[BEGIN FINAL RESPONSE]",
+    #     "OUTPUT_TAG_END": "",
+    #     "OUTPUT_STOP_TOKEN": "<|end|>"
+    # },
     "Apriel-1.5-15B-thinker": {
         "MODEL_DISPLAY_NAME": "Apriel-1.5-15B-thinker",
         "MODEL_HF_URL": "https://huggingface.co/ServiceNow-AI/Apriel-1.5-15b-Thinker",
@@ -21,7 +35,10 @@ models_config = {
         "AUTH_TOKEN": os.environ.get("AUTH_TOKEN"),
         "REASONING": True,
         "MULTIMODAL": True,
-        "TEMPERATURE": 0.6
+        "TEMPERATURE": 0.6,
+        "OUTPUT_TAG_START": "[BEGIN FINAL RESPONSE]",
+        "OUTPUT_TAG_END": "[END FINAL RESPONSE]",
+        "OUTPUT_STOP_TOKEN": "<|end|>"
     },
     # "Apriel-Nemotron-15b-Thinker": {
     #     "MODEL_DISPLAY_NAME": "Apriel-Nemotron-15b-Thinker",